Short-term exposure to PM2.5 and 1.5 million deaths: a time-stratified case-crossover analysis in the Mexico City Metropolitan Area.

BACKGROUND: Satellite-based PM2.5 predictions are being used to advance exposure science and air-pollution epidemiology in developed countries; including emerging evidence about the impacts of PM2.5 on acute health outcomes beyond the cardiovascular and respiratory systems, and the potential modifying effects from individual-level factors in these associations. Research on these topics is lacking in low and middle income countries. We aimed to explore the association between short-term exposure to PM2.5 with broad-category and cause-specific mortality outcomes in the Mexico City Metropolitan Area (MCMA), and potential effect modification by age, sex, and SES characteristics in such associations. METHODS: We used a time-stratified case-crossover study design with 1,479,950 non-accidental deaths from the MCMA for the period of 2004-2019. Daily 1 × 1 km PM2.5 (median = 23.4 µg/m3; IQR = 13.6 µg/m3) estimates from our satellite-based regional model were employed for exposure assessment at the sub-municipality level. Associations between PM2.5 with broad-category (organ-system) and cause-specific mortality outcomes were estimated with distributed lag conditional logistic models. We also fit models stratifying by potential individual-level effect modifiers including; age, sex, and individual SES-related characteristics namely: education, health insurance coverage, and job categories. Odds ratios were converted into percent increase for ease of interpretation. RESULTS: PM2.5 exposure was associated with broad-category mortality outcomes, including all non-accidental, cardiovascular, cerebrovascular, respiratory, and digestive mortality. A 10-µg/m3 PM2.5 higher cumulative exposure over one week (lag06) was associated with higher cause-specific mortality outcomes including hypertensive disease [2.28% (95%CI: 0.26%-4.33%)], acute ischemic heart disease [1.61% (95%CI: 0.59%-2.64%)], other forms of heart disease [2.39% (95%CI: -0.35%-5.20%)], hemorrhagic stroke [3.63% (95%CI: 0.79%-6.55%)], influenza and pneumonia [4.91% (95%CI: 2.84%-7.02%)], chronic respiratory disease [2.49% (95%CI: 0.71%-4.31%)], diseases of the liver [1.85% (95%CI: 0.31%-3.41%)], and renal failure [3.48% (95%CI: 0.79%-6.24%)]. No differences in effect size of associations were observed between age, sex and SES strata. CONCLUSIONS: Exposure to PM2.5 was associated with non-accidental, broad-category and cause-specific mortality outcomes beyond the cardiovascular and respiratory systems, including specific death-causes from the digestive and genitourinary systems, with no indication of effect modification by individual-level characteristics.

Can weather help explain 'why now?': The potential role of hourly temperature as a stroke trigger.

BACKGROUND: While evidence suggests that daily ambient temperature exposure influences stroke risk, little is known about the potential triggering role of ultra short-term temperature. METHODS: We examined the association between hourly temperature and ischemic and hemorrhagic stroke, separately, and identified any relevant lags of exposure among adult New York State residents from 2000 to 2015. Cases were identified via ICD-9 codes from the New York Department of Health Statewide Planning and Reearch Cooperative System. We estimated ambient temperature up to 36 h prior to estimated stroke onset based on patient residential ZIP Code. We applied a time-stratified case-crossover study design; control periods were matched to case periods by year, month, day of week, and hour of day. Additionally, we assessed effect modification by leading stroke risk factors hypertension and atrial fibrillation. RESULTS: We observed 578,181 ischemic and 164,755 hemorrhagic strokes. Among ischemic and hemorrhagic strokes respectively, the mean (standard deviation; SD) patient age was 71.8 (14.6) and 66.8 (17.4) years, with 55% and 49% female. Temperature ranged from -29.5 °C to 39.2 °C, with mean (SD) 10.9 °C (10.3 °C). We found linear relationships for both stroke types. Higher temperature was associated with ischemic stroke over the 7 h following exposure; a 10 °C increase over 7 h was associated with 5.1% (95% Confidence Interval [CI]: 3.8, 6.4%) increase in hourly stroke rate. In contrast, temperature was negatively associated with hemorrhagic stroke over 5 h, with a 5-h cumulative association of -6.2% (95% CI: 8.6, -3.7%). We observed suggestive evidence of a larger association with hemorrhagic stroke among patients with hypertension and a smaller association with ischemic stroke among those with atrial fibrillation. CONCLUSION: Hourly temperature was positively associated with ischemic stroke and negatively associated with hemorrhagic stroke. Our results suggest that ultra short-term weather influences stroke risk and hypertension may confer vulnerability.

The association between ambient temperature variability and myocardial infarction in a New York-State-based case-crossover study: An examination of different variability metrics.

BACKGROUND: Short-term temperature variability has been consistently associated with mortality, with limited evidence for cardiovascular outcomes. Previous studies have used multiple metrics to measure temperature variability; however, those metrics do not capture hour-to-hour changes in temperature. OBJECTIVES: We assessed the correlation between sub-daily temperature-change-over-time metrics and previously-used metrics, and estimated associations with myocardial infarction (MI) hospitalizations. METHODS: Hour-to-hour change-over-time was measured via three metrics: 24-hr mean absolute hourly first difference, 24-hr maximum absolute hourly first difference, and 24-hr mean hourly first difference. We first assessed the Spearman correlations between these metrics and four previously-used metrics (24-hr standard deviation of hourly temperature, 24-hr diurnal temperature range, 48-hr standard deviation of daily minimal and maximal temperatures, and 48-hr difference of daily mean temperature), using hourly data from the North America Land Data Assimilation System-2 Model. Subsequently, we estimated the association between these metrics and primary MI hospitalization in adult residents of New York State for 2000-2015 using a time-stratified case-crossover design. RESULTS: The hour-to-hour change-over-time metrics were correlated, but not synonymous, with previously-used metrics. We observed 809,259 MI, 45% of which were among females and the mean (standard deviation) age was 70 (15). An increase from mean to 90th percentile in mean absolute first difference of temperature was associated with a 2.04% (95% Confidence Interval [CI]: 1.30-2.78%) increase in MI rate. An increase from mean to 90th percentile in mean first difference also yielded a positive association (1.86%; 95%CI: 1.09-2.64%). We observed smaller- or similar-in-magnitude positive associations for previously-used metrics. DISCUSSION: First, short-term hour-to-hour temperature change was positively associated with MI risk. Second, all other variability metrics yielded positive associations with MI, with varying magnitude. In future research on temperature variability, researchers should define their research question, including which aspects of variability they intend to measure, and apply the appropriate metric. ALTERNATIVE: All metrics of temperature variability, including short-term hour-to-hour temperature changes, were positively associated with MI risk, though the magnitude of effect estimates varied by metric.

A 1-km hourly air-temperature model for 13 northeastern U.S. states using remotely sensed and ground-based measurements.

BACKGROUND: Accurate and precise estimates of ambient air temperatures that can capture fine-scale within-day variability are necessary for studies of air temperature and health. METHOD: We developed statistical models to predict temperature at each hour in each cell of a 927-m square grid across the Northeast and Mid-Atlantic United States from 2003 to 2019, across ~4000 meteorological stations from the Integrated Mesonet, using inputs such as elevation, an inverse-distance-weighted interpolation of temperature, and satellite-based vegetation and land surface temperature. We used a rigorous spatial cross-validation scheme and spatially weighted the errors to estimate how well model predictions would generalize to new cell-days. We assess the within-county association of temperature and social vulnerability in a heat wave as an example application. RESULTS: We found that a model based on the XGBoost machine-learning algorithm was fast and accurate, obtaining weighted root mean square errors (RMSEs) around 1.6 K, compared to standard deviations around 11.0 K. We found similar accuracy when validating our model on an external dataset from Weather Underground. Assessing predictions from the North American Land Data Assimilation System-2 (NLDAS-2), another hourly model, in the same way, we found it was much less accurate, with RMSEs around 2.5 K. This is likely due to the NLDAS-2 model's coarser spatial resolution, and the dynamic variability of temperature within its grid cells. Finally, we demonstrated the health relevance of our model by showing that our temperature estimates were associated with social vulnerability across the region during a heat wave, whereas the NLDAS-2 showed a much weaker association. CONCLUSION: Our high spatiotemporal resolution air temperature model provides a strong contribution for future health studies in this region.

Advancing methodologies for applying machine learning and evaluating spatiotemporal models of fine particulate matter (PM2.5) using satellite data over large regions.

Reconstructing the distribution of fine particulate matter (PM2.5) in space and time, even far from ground monitoring sites, is an important exposure science contribution to epidemiologic analyses of PM2.5 health impacts. Flexible statistical methods for prediction have demonstrated the integration of satellite observations with other predictors, yet these algorithms are susceptible to overfitting the spatiotemporal structure of the training datasets. We present a new approach for predicting PM2.5 using machine-learning methods and evaluating prediction models for the goal of making predictions where they were not previously available. We apply extreme gradient boosting (XGBoost) modeling to predict daily PM2.5 on a 1×1 km2 resolution for a 13 state region in the Northeastern USA for the years 2000-2015 using satellite-derived aerosol optical depth and implement a recursive feature selection to develop a parsimonious model. We demonstrate excellent predictions of withheld observations but also contrast an RMSE of 3.11 µg/m3 in our spatial cross-validation withholding nearby sites versus an overfit RMSE of 2.10 µg/m3 using a more conventional random ten-fold splitting of the dataset. As the field of exposure science moves forward with the use of advanced machine-learning approaches for spatiotemporal modeling of air pollutants, our results show the importance of addressing data leakage in training, overfitting to spatiotemporal structure, and the impact of the predominance of ground monitoring sites in dense urban sub-networks on model evaluation. The strengths of our resultant modeling approach for exposure in epidemiologic studies of PM2.5 include improved efficiency, parsimony, and interpretability with robust validation while still accommodating complex spatiotemporal relationships.

Residential segregation and summertime air temperature across 13 northeastern U.S. states: Potential implications for energy burden.

High ambient summertime temperatures are an increasing health concern with climate change. This is a particular concern for minoritized households in the United States, for which differential energy burden may compromise adaptive capacity to high temperatures. Our research question was: Do minoritized groups experience hotter summers than the area average, and do non-Hispanic white people experience cooler summers? Using a fine-scaled spatiotemporal air temperature model and U.S. census data, we examined local (within-county) differences in warm season cooling degree days (CDDs) by ethnoracial group as a proxy for local energy demand for space cooling across states of the northeast and mid-Atlantic U.S. in 2003-2019. Using state-specific regression models adjusted for year and county, we found that Black and Latino people consistently experienced more CDDs, non-Hispanic white people experienced fewer CDDs, and Asian populations showed mixed results. We also explored a concentration-based measure of residential segregation for each ethnoracial group as one possible pathway towards temperature disparities. We included the segregation measure as a smooth term in a regression model adjusted for county and year. The results were nonlinear, but higher concentrations of white people were associated with lower annual CDDs and higher concentrations of Latino people were associated with higher annual CDDs than the county average. Concentrations for Black and Asian people were nonmonotonic, sometimes with bowed associations. These findings suggest that present-day residential segregation, as modeled by spatially smoothed ethnoracial subgroup concentrations, may contribute to summertime air temperature disparities and influence adaptive capacity. We hope these findings can support place-based interventions, including targeting of energy insecurity relief programs.

Daily exposure to PM2.5 and 1.5 million deaths: A time-stratified case-crossover analysis in the Mexico City Metropolitan Area.

Background: Satellite-based PM2.5 predictions are being used to advance exposure science and air-pollution epidemiology in developed countries; including emerging evidence about the impacts of PM2.5 on acute health outcomes beyond the cardiovascular and respiratory systems, and the potential modifying effects from individual-level factors in these associations. Research on these topics is lacking in Latin America. Methods: We used a time-stratified case-crossover study design with 1,479,950 non-accidental deaths from Mexico City Metropolitan Area for the period of 2004-2019. Daily 1×1 km PM2.5 (median=23.4 µg/m3; IQR=13.6 µg/m3) estimates from our satellite-based regional model were employed for exposure assessment at the sub-municipality level. Associations between PM2.5 with broad-category (organ-system) and cause-specific mortality outcomes were estimated with distributed lag conditional logistic models. We also fit models stratifying by potential individual-level effect modifiers including; age, sex, and individual SES-related characteristics namely: education, health insurance coverage, and job categories. Results: PM2.5 exposure was associated with higher total non-accidental, cardiovascular, cerebrovascular, respiratory, and digestive mortality. A 10-µg/m3 PM2.5 higher cumulative exposure over one week (lag06) was associated with higher cause-specific mortality outcomes including hypertensive disease [2.28% (95%CI: 0.26%-4.33%)], acute ischemic heart disease [1.61% (95%CI: 0.59%-2.64%)], other forms of heart disease [2.39% (95%CI: -0.35%-5.20%)], hemorrhagic stroke [3.63% (95%CI: 0.79%-6.55%)], influenza and pneumonia [4.91% (95%CI: 2.84%-7.02%)], chronic respiratory disease [2.49% (95%CI: 0.71%-4.31%)], diseases of the liver [1.85% (95%CI: 0.31%-3.41%)], and renal failure [3.48% (95%CI: 0.79%-6.24%)]. No differences in effect size of associations were observed between SES strata. Conclusions: Exposure to PM2.5 was associated with mortality outcomes beyond the cardiovascular and respiratory systems, including specific death-causes from the digestive and genitourinary systems, with no indications of effect modification by individual SES-related characteristics.

Prediction of daily mean and one-hour maximum PM2.5 concentrations and applications in Central Mexico using satellite-based machine-learning models.

BACKGROUND: Machine-learning algorithms are becoming popular techniques to predict ambient air PM2.5 concentrations at high spatial resolutions (1 × 1 km) using satellite-based aerosol optical depth (AOD). Most machine-learning models have aimed to predict 24 h-averaged PM2.5 concentrations (mean PM2.5) in high-income regions. Over Mexico, none have been developed to predict subdaily peak levels, such as the maximum daily 1-h concentration (max PM2.5). OBJECTIVE: Our goal was to develop a machine-learning model to predict mean PM2.5 and max PM2.5 concentrations in the Mexico City Metropolitan Area from 2004 through 2019. METHODS: We present a new modeling approach based on extreme gradient boosting (XGBoost) and inverse-distance weighting that uses AOD, meteorology, and land-use variables. We also investigated applications of our mean PM2.5 predictions that can aid local authorities in air-quality management and public-health surveillance, such as the co-occurrence of high PM2.5 and heat, compliance with local air-quality standards, and the relationship of PM2.5 exposure with social marginalization. RESULTS: Our models for mean and max PM2.5 exhibited good performance, with overall cross-validated mean absolute errors (MAE) of 3.68 and 9.20 µg/m3, respectively, compared to mean absolute deviations from the median (MAD) of 8.55 and 15.64 µg/m3. In 2010, everybody in the study region was exposed to unhealthy levels of PM2.5. Hotter days had greater PM2.5 concentrations. Finally, we found similar exposure to PM2.5 across levels of social marginalization. SIGNIFICANCE: Machine learning algorithms can be used to predict highly spatiotemporally resolved PM2.5 concentrations even in regions with sparse monitoring. IMPACT: Our PM2.5 predictions can aid local authorities in air-quality management and public-health surveillance, and they can advance epidemiological research in Central Mexico with state-of-the-art exposure assessment methods.

A spatiotemporal reconstruction of daily ambient temperature using satellite data in the Megalopolis of Central Mexico from 2003 to 2019.

While weather stations generally capture near-surface ambient air temperature (Ta) at a high temporal resolution to calculate daily values (i.e., daily minimum, mean, and maximum Ta), their fixed locations can limit their spatial coverage and resolution even in densely populated urban areas. As a result, data from weather stations alone may be inadequate for Ta-related epidemiology particularly when the stations are not located in the areas of interest for human exposure assessment. To address this limitation in the Megalopolis of Central Mexico (MCM), we developed the first spatiotemporally resolved hybrid satellite-based land use regression Ta model for the region, home to nearly 30 million people and includes Mexico City and seven more metropolitan areas. Our model predicted daily minimum, mean, and maximum Ta for the years 2003-2019. We used data from 120 weather stations and Land Surface Temperature (LST) data from NASA's MODIS instruments on the Aqua and Terra satellites on a 1 × 1 km grid. We generated a satellite-hybrid mixed-effects model for each year, regressing Ta measurements against land use terms, day-specific random intercepts, and fixed and random LST slopes. We assessed model performance using 10-fold cross-validation at withheld stations. Across all years, the root-mean-square error ranged from 0.92 to 1.92 K and the R 2 ranged from .78 to .95. To demonstrate the utility of our model for health research, we evaluated the total number of days in the year 2010 when residents ≥65 years old were exposed to Ta extremes (above 30°C or below 5°C). Our model provides much needed high-quality Ta estimates for epidemiology studies in the MCM region.